GitHub Enterprise Server 3.23 release candidate
GitHub released Enterprise Server 3.23 with CPU-based air-gapped secret scanning, eight-hour OAuth tokens, and OpenSSF malware alerts across seven ecosystems.
By Dillip Chowdary • Nov 20, 2026 • Source: GitHub Changelog

At a glance
- Product
- GitHub Enterprise Server
- Version
- 3.23 release candidate
- Availability
- Public preview for AI secrets and expiring OAuth
- Token duration
- 8 hours with companion refresh tokens
- Dependabot delay
- 3 days on package registries before pull requests
- Hardware requirements
- Appliance CPU without GPU or outbound connection
In this post
GitHub launched the release candidate for GitHub Enterprise Server 3.23, introducing administrative upgrades, enterprise workflow enhancements, and security controls across on-premises infrastructure. According to GitHub Changelog's report, this release candidate allows organizations to test new operational capabilities early and provide feedback prior to general availability. The update spans native project planning, local artificial intelligence secret scanning, ecosystem malware detection, expiring authentication tokens, and cluster database topology.
This report details the architectural changes, operational parameters, and deployment prerequisites in GitHub Enterprise Server 3.23. The coverage is designed for enterprise systems administrators, security engineers, and platform developers who manage self-hosted GitHub deployments or build enterprise OAuth integrations. Readers will find exact timing rules, supported ecosystems, credential lifecycles, and cluster node isolation mechanics.
01Complex project views with boolean filters
Teams organizing issues and pull requests inside GitHub Projects can now construct precise project views using compound logic. The upgraded filtering engine accepts logical OR conditions, explicit AND parameters, and nested parentheses for grouping multiple terms together. For instance, filtering by (status:Done OR status:Todo) is:issue isolates issues in either stage while omitting intermediate or untracked work items.
Before this capability arrived, filtering across diverse sprint boards required maintaining separate views or relying on restricted query syntax. Project managers and team leads can now consolidate multiple task states into unified views without losing track of priority items. The syntax processes grouping rules consistently so cross-functional engineering teams can evaluate planning boards with custom query precision.
02On-premises artificial intelligence secret detection
Enterprise environments running GitHub Secret Protection can now identify unstructured secrets entirely within their local network perimeters. The new AI secret detection system functions in completely air-gapped environments without initiating any outbound connections to hosted large language models. The architecture executes a self-contained, fine-tuned model directly on the server appliance CPU, eliminating any requirement for specialized graphics processing units.
Running detection routines on CPU hardware avoids the infrastructure costs and hardware procurement hurdles associated with GPU hosting in private data centers. Security personnel can detect unstructured credentials and sensitive strings embedded within proprietary repositories while adhering to strict internal governance standards. This processing model ensures that proprietary source code and detected tokens never leave the customer network boundary during analysis.
Note: Artificial intelligence secret detection on GitHub Enterprise Server is currently in public preview and remains subject to change. Organizations testing this capability should evaluate detection performance across internal codebases and report feedback directly to GitHub Support before general deployment across production enterprise server clusters.
03Ecosystem malware alerts and update cooldowns
The GitHub Advisory Database has expanded its security threat monitoring by identifying known malicious packages across PyPI, Maven, RubyGems, NuGet, Go, Rust, and Composer. This broadening builds upon existing npm surveillance by automatically ingesting malware advisories from OpenSSF's malicious-packages repository. Platform engineers and software developers gain automated awareness when upstream package registries publish libraries identified as malicious by the open-source security community.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
To further harden software supply chains, Dependabot now introduces an automated three-day quarantine window for package version updates. By default, Dependabot waits until a new package version has been published on its registry for at least three full days before opening an automated update pull request. This intentional pause grants package maintainers and security researchers critical time to detect, report, and remove compromised or defective releases before enterprise teams merge them.
| Ecosystem or Feature | Previous GHES Baseline | GHES 3.23 Release Candidate |
|---|---|---|
| Malware Advisories | npm coverage only | PyPI, Maven, RubyGems, NuGet, Go, Rust, Composer, npm |
| Advisory Ingestion | Standard security feeds | Automated OpenSSF malicious-packages repository ingestion |
| Dependabot Cooldown | Zero delay after registry publish | Mandatory three-day waiting window before pull requests |
| Secret Scanning AI | External network or GPU needed | Local appliance CPU execution without outbound connections |
The combination of open-source advisory data and delayed pull requests establishes layered defense against zero-day supply chain compromises. Engineering organizations receive protection across eight language ecosystems while avoiding immediate adoption of unverified software releases. These automated controls reduce the risk of integrating malicious dependencies into enterprise build pipelines.
04Expiring OAuth tokens and refresh credentials
Application developers can now restrict credential exposure by generating OAuth access tokens that expire automatically after eight hours. To maintain continuous application access without manual reauthentication, applications use companion refresh tokens to obtain new access credentials. Token expiration is enabled by default for all newly registered OAuth applications, while existing OAuth registrations can opt into the expiration policy.
Developers testing this authentication lifecycle can request expiring tokens by including scope=offline_access during authorization requests to facilitate gradual rollouts. Alternatively, administrators can enforce expiration directly within their application settings once client software is updated. Engineering teams must ensure that all external client software correctly requests, stores, and exchanges refresh tokens before enforcing mandatory expiration, preventing authentication failures for legacy tools.
Note: Expiring OAuth access tokens and refresh token functionality are currently in public preview and subject to operational change. Teams should verify refresh token exchange logic across all active client installations before requiring expiration in OAuth application registrations.
05Dedicated Redis primary nodes in clusters
Cluster administrators running GitHub Enterprise Server can now isolate the Redis primary instance onto a dedicated cluster node. Previously, Redis shared compute, memory, and storage resources with other cluster services such as MySQL on the same server instance. Under the updated architecture, Redis uses the dedicated node's computing resources and independent data volume rather than competing with database workloads.
This architectural migration reallocates service placement across cluster infrastructure without attaching new storage disks to existing nodes or routing traffic to external Redis services. Isolating the primary cache and key-value store improves resource predictability and prevents database bottlenecks during periods of heavy enterprise activity. System administrators can execute this node relocation while maintaining established cluster boundary constraints and storage layouts.
Organizations planning to evaluate these infrastructure capabilities can download the GitHub Enterprise Server 3.23 release candidate directly from GitHub. Release candidates allow enterprise platform teams to test upgrades against non-production staging environments and consult the GHES 3.23 release notes before general availability. Questions, operational defects, and migration feedback can be submitted directly through official GitHub Support channels.
Frequently asked questions
How does AI secret detection operate in air-gapped environments?
A self-contained, fine-tuned model runs directly on the appliance CPU. It requires no GPU, hosted LLM, or outbound network connection.
Which package ecosystems now have malware advisory coverage?
The GitHub Advisory Database covers PyPI, Maven, RubyGems, NuGet, Go, Rust, and Composer alongside npm. It ingests advisories from OpenSSF's malicious-packages repository.
How long does Dependabot wait before submitting version updates?
Dependabot pauses for at least three days after a package version appears on its registry before opening an update pull request. This cooldown allows maintainers to identify and report compromised releases.
How do token lifetimes change for OAuth applications?
Newly created OAuth apps default to access tokens that expire after eight hours and use refresh tokens. Existing applications can opt in or request expiring tokens using scope=offline_access.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement